Uday Paripelli

Architected and deployed LLM-powered agentic systems using LangChain, LangGraph, and CrewAI to enable inter-agent memory and collaboration, improving workflow task completion by 30% across multiple financial operations. Fine-tuned LLMs (Hugging Face, LangChain) for clinical and legal summarization tasks, increasing accuracy by 27% and reducing analyst review workload by ~15 hours per week. Deployed Generative AI solutions via AWS Bedrock (Claude, GPT-4) for document summarization and regulatory filings, achieving ~45% faster processing with scalable automation. Optimized SageMaker lifecycle configurations and right-sizing strategies, cutting compute costs by ~18% (~$200K annually). Designed and implemented RAG pipelines with Pinecone and FAISS, improving document retrieval accuracy by ~35% and saving ~$400K yearly. Built a real-time fraud detection pipeline using XGBoost, PySpark, and SageMaker, achieving sub-second predictions, 42% higher detection accuracy, and 30% fewer false positi

Conducted A/B testing of LLM prompts using LangChain and evaluation metrics (BLEU, ROUGE), improving chatbot response accuracy by 18% and customer satisfaction scores. Built deep learning models (CNN, RNN, LSTM, Transformers) for anomaly detection and predictive maintenance, achieving >90% accuracy and reducing MTTR by 25%. Automated ETL pipelines using Python, SQL, and SAS to streamline model ingestion, improving data throughput by 60% and reducing preparation time from days to hours. Developed scalable ML pipelines using PySpark, AWS Glue, and Redshift for real-time analytics, improving data delivery speed and business insights by 60%. Designed XGBoost-based failure prediction models that reduced system downtime by 15% and prevented ~$1.2M in annual revenue loss. Implemented MLOps pipelines with Docker, Kubernetes, and MLflow to deploy and monitor financial models, achieving 99.9% uptime and reducing model release cycles by 35%. Built interactive dashboards and EDA visualizations (Ma

Built deep learning models for anomaly detection and predictive maintenance achieving >90% accuracy; automated ETL pipelines; implemented MLOps pipelines with Docker, Kubernetes, and MLflow to deploy and monitor models with 99.9% uptime; led data analytics and dashboards.